Model discrimination in chemical kinetics

1876-214X/09 2009 Bentham Open

Open Access

Model Discrimination in Chemical Kinetics

Burcu Özdemir and Selahattin Gültekin

Dou University, Acıbadem, Kadıköy 34722, Istanbul, Turkey

Abstract: In studies on chemical kinetics, generally after the rate data have been taken, a mechanism and an associated

rate law model are proposed based on the data taken. Frequently, more than one mechanism and rate law may be consis-tent with data. In order to find the correct rate law, regression techniques (model discrimination) are applied to indentify which model equation best fits the data by choosing the one with the smaller sum of squares. With this non-linear regres-sion technique, rate parameters with 95% confidence limits are calculated along with residues. Of course, model parame-ters must be realistic. For example, reaction rate constants, activation energies or adsorption equilibrium constants must be positive by comparing the calculated value of parameters with 95% confidence limits, one can judge about the validity of the model.

In this paper, model discrimination will be applied to certain data from the literature along with the suggested heterogene-ous catalytic models such as Langmuir-Hinshelwood Kinetic Model or Rideal-Eley Model.

A reaction of A+ B
C + D type have been selected (like methanation) with the rate laws given below:

rA = kPAPB / 1+ K

(

)

2

dual-site Langmuir-Hinshelwood Model
rA = kPAPB / 1

(

)

rA = kPAPB / 1+ K

(

)

2

only reactants are adsorbed
rA = kPAPB / 1+ K

(

)

2

dual-site Rideal-Eley Model
rA = kPA a_{ P} B b _power-law In this study, CO+ 3H2
CH4+ H2O

reaction rate data were tested for five different models to find the most suitable rate expression by model discrimination method taking the advantage of powerful POLYMATH package program.

INTRODUCTION Regression Techniques

In empirical studies, in order to determine the parameters for the postulated model, very powerful regression tech-niques (methods) are used.

Regression methods basically are as follows [1]

a) Linear regression (like y= ax + b) , where a and b are to be determined.

b) Multiple regression (likey= a₁x₁+ a₂x₂+ ....+ a_nx_n), where ai’s are parameters to be determined.

c) Polynomial regression (like

y= a_nxn_{+ a}

n
1xn
1+ ...+ a1x+ a0), ai’s are parameter

to be determined from regression.

*Address correspondence to this author at the Dou University, Acıbadem, Kadıköy 34722, Istanbul, Turkey; E-mail: sgultekin@dogus.edu.tr

d) Non-linear regression [2].

This is very common and can be used almost under any condition. General form is

y= f (x₁, x₂,..., x_n, a₁, a₂,..., a_n), where n = # of ex-periments, m = # of parameters to be determined pro-viding n> m+1.

The common features of the above regression techniques are that it is to make the variance minimum, and to make the correction factor as close to unity as possible. After determi-nation of parameters, the next step is to check to see whether they are physically meaningful or not. For example, if ad-sorption equilibrium constant is one of the parameters, then it must decrease with increasing temperature as adsorption is an exothermic process (Le Chatelier Principle)

In kinetic studies, generally one faces very complicated rate expressions on heterogeneous catalysts. Those rate ex-pressions may obey Langmuir-Hinshelwood or Rideal-Eley model [3, 4].

In general, if we have a reaction of A+ B
Cwe may have possible rate expressions as

2 The Open Catalysis Journal, 2009, Volume 2 Özdemir and Gültekin
rA = kPAPB (1+ K_AP_A+ K_BP_B+ K_CP_C)2 (Dual-site model)
rA= kP_AP_B (1+ KAPA+ KBPB+ KCPC (Single-site model)
rA = kP_AP_B (1+ K_AP_A)2 (Dual-site, Rideal-Eley)
rA= kPA a PB

b _{(Apparent power-law expression)}

In kinetic studies, for example, one may have three dif-ferent mechanisms and three difdif-ferent rate-determining steps. Therefore, one will have nine different rate expressions. In order to determine the correct rate expression, model

dis-crimination method is being used. The essence of this

method is not only to minimize the variance, but also to keep the correction factor as close to one (unity) as possible. In addition to the above two criteria, one also has to check the physical validity of the determined parameters.

RESULTS

Now we will chose the reaction of methanation [4] on Ni catalyst where initial rates are taken at constant temperature at variety partial pressures of reactants and products.

Consider the reaction

CO+ 3H2
CH4+ H2O (methanation)

Assuming the following runs were carried out under the given conditions (see Table 1).

Table 1. Initial Rates Obtained at Various Partial Pressures

Run # rA* PCO** PH2** PH2O** PCH4** 1 0.1219 1 1 0 0 2 0.0944 1 1 1 1 3 0.0943 1 1 1 2 4 0.0753 1 1 2 1 5 0.0753 1 1 2 2 6 0.0512 1 1 4 1 7 0.0280 1 1 8 1 8 0.1274 1 2 1 1 9 0.1056 1 2 2 2 10 0.1203 1 4 2 2 11 0.1189 1 8 1 1 12 0.0782 2 1 1 1 13 0.1204 2 2 1 1 14 0.1057 2 2 2 0 15 0.1056 2 2 2 1 16 0.1056 2 2 2 2 17 0.1552 2 4 1 1 18 0.0533 4 1 1 1 19 0.0911 4 2 1 1 20 0.0317 8 1 1 1 21 0.1476 8 8 1 1 * rA [mole/kg.cat.s], ** Pi [atm].

We will consider the following plausible rate expressions and by model discrimination we will reach hopefully to the true rate expression.

a) All reactants and products are adsorbed with dual-site mechanism:
rA= kPCO.PH2 (1+ KCO.PCO+ KH2.PH2 + KH2O.PH2O+ KCH4PCH4) 2 (Dual-site)

b) All reactants and products are adsorbed with single-site mechanism:
rA= kP_CO.P_H 2 (1+ K_CO.P_CO+ K_H 2.PH2 + KH2O.PH2O+ KCH4PCH4) (Single-site)

c) Only reactants are adsorbed with dual-site mecha-nism:
rA = kP_CO.P_H 2 (1+ K_CO.P_CO+ K_H 2PH2)

2 (Only reactants are

adsorbed, dual-site)

d) Only CO is adsorbed; H2 comes directly from gas

phase and reacts with adsorbed CO (Rideal-Eley Mechanism)
rA = kP_CO.P_H 2 (1+ K_CO.P_CO)2 (Rideal-Eley, dual-site) e) A power-law expression:
rA = k.PCO a _.P H2 b _(Power-Law)

In the following pages, these five different models were tried on Polymath® _{program. The results are self explanatory,}

and are given in Figs. (1-5).

Model a:
rA = kP_CO.P_H 2 (1+ K_CO.P_CO+ K_H 2.PH2+ KH2O.PH2O+ KCH4PCH4) 2 Output of Model a: rA = kPCOPH / 1+ K

(

)

Variable Ini. Guess Value 95% Confidence

k 4 7.9628870 0.0367133

KCO 2 5.0877643 0.0124837 KH 4 1.9950241 0.0047754 KH2O 5 1.0969188 0.0030085

KCH4 1 0.0046140 0.0015179

Nonlinear regression settings. Max # iterations = 64

Fig. (1). Input of the data in POLYMATH Screen for Model a.

4 The Open Catalysis Journal, 2009, Volume 2 Özdemir and Gültekin Precision R^2 = 0.9999995 R^2adj = 0.9999993 Rmsd = 5.321E-06 Variance = 7.804E-10 Model b:
r_A = kPCO.PH2 (1+ KCO.PCO+ KH2.PH2 + KH2O.PH2O+ KCH4PCH4) Output of Model b: rA = kPCOPH / 1

(

)

Variable Ini guess Value 95% Confidence k 14.0 2.747922 0.0023620 KCO 8.0 51.41168 0.0593556

KH 1.0 12.89503 0.0368865 KH2O 1.0 2.123223 0.0692380

KCH4 0.01 -1.989895 0.0982922

Nonlinear regression settings Max # iterations = 64 Precision R^2 = -1.586834 R^2adj = -2.233542 Rmsd = 0.011789 Variance = 0.0038306 Model c:
r_A = kPCO.PH2 (1+ K_CO.P_CO+ K_H 2PH2) 2 Output of Model c: r_A= kP_COP_H / 1

(

)

k 14.0 0.7700806 0.6088063 KCO 8.0 1.540737 0.7859312 KH 1.0 0.5896616 0.3142481 Precision R^2 = 0.7311287 R^2adj = 0.7012541 Rmsd = 0.0038007 Variance = 0.0003539 Model d:
r_A = kPCO.PH2 (1+ K_CO.P_CO)2

Output of Model d: rA= kPCOPH / 1+ K

(

)

2

Variable Ini Guess Value 95% Confidence k 14.0 0.062234 5.47E-06 KCO 8.0 0.453103 3.977E-05

Nonlinear regression settings. Max # iterations = 64 Precision R^2 = -1.282797 R^2adj = -1.402944 Rmsd = 0.0110745 Variance = 0.0028467 Model e:
rA = k.PCO a _.P H2 b

Output of Model e: ra = k*PCO^a*PH^b

Fig. (4). Input of the data in POLYMATH Screen for Model d.

6 The Open Catalysis Journal, 2009, Volume 2 Özdemir and Gültekin Variable Ini Guess Value 95% Confidence

k 14.0 0.0806184 3.076E-06 a 1.0 -0.0554411 4.271E-05 b 1.0 0.3205555 3.242E-05

Nonlinear regression settings. Max # iterations = 64 Precision R^2 = 0.4902824 R^2adj = 0.4336471 Rmsd = 0.0052331 Variance = 6.709E-04

When variance, correction factor (coefficient) and the physical meaningfulness of the parameters are considered, one can clearly see that the model a fits the data in a perfect manner. In the remaining other models (i.e. b, c, d and e) either because of variance or correction factor or 95% confi-dence interval or combination thereof, the models seem to be inadequate and as a result they are eliminated.

CONCLUSIONS

No matter how the chemical kinetic expressions are complicated, either linear, multiple linear, non-linear or polynomial, one can find a satisfactory rate expression by means of model discrimination method. This discrimination processes are eased especially after the advent of powerful ready package programs such as POLYMATH, MATLAB, MATHCAD [5], and others. One, still, has to be very cau-tious in that finding the satisfactory rate expression does not mean the true (absolute) rate expression and mechanism are found [6].

ACKNOWLEDGEMENTS

Authors would like to express their thanks to Dou University for its financial support.

REFERENCES

[1] Polymath 6.1 User Guide, 2006.

[2] MATLAB R2006b Mathworks’ Users Manual, 2006.

[3] Fogler, H.S. Elements of Chemical Reaction Engineering, 4/E, Prentice-Hall, 2006.

[4] Satterfield, C.N. Industrial Heterogeneous Catalytic Processes, 2/E, McGraw-Hill, 1990.

[5] Gültekin, S. Kimya Mühendisliinde MATHCAD Kullanımı (Usage of MATHCAD in Chemical Engineering), Lecture Notes, Yıldız Technical University, 1997.

[6] Cutlip, M.B.; Shacham, M. Problem Solving in Chemical and Biochemical Engineering with POLYMATH, Excel and MAT-LAB, 2/E, Prentice-Hall, 2008.

Received: December 26, 2008 Revised: December 31, 2008 Accepted: December 31, 2008

© Özdemir and Gültekin; Licensee Bentham Open.

This is an open access article licensed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted, non-commercial use, distribution and reproduction in any medium, provided the work is properly cited.